db/btree/FreeBSD/bt_open.c

   1 /*-
   2  * Copyright (c) 1990, 1993, 1994
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to Berkeley by
   6  * Mike Olson.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  * 3. All advertising materials mentioning features or use of this software
  17  *    must display the following acknowledgement:
  18  *      This product includes software developed by the University of
  19  *      California, Berkeley and its contributors.
  20  * 4. Neither the name of the University nor the names of its contributors
  21  *    may be used to endorse or promote products derived from this software
  22  *    without specific prior written permission.
  23  *
  24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  34  * SUCH DAMAGE.
  35  */
  36
  37 #if defined(LIBC_SCCS) && !defined(lint)
  38 static char sccsid[] = "@(#)bt_open.c   8.10 (Berkeley) 8/17/94";
  39 #endif /* LIBC_SCCS and not lint */
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD: src/lib/libc/db/btree/bt_open.c,v 1.11 2002/03/22 21:52:01 obrien Exp $");
  42
  43 /*
  44  * Implementation of btree access method for 4.4BSD.
  45  *
  46  * The design here was originally based on that of the btree access method
  47  * used in the Postgres database system at UC Berkeley.  This implementation
  48  * is wholly independent of the Postgres code.
  49  */
  50
  51 #include "namespace.h"
  52 #include <sys/param.h>
  53 #include <sys/stat.h>
  54
  55 #include <errno.h>
  56 #include <fcntl.h>
  57 #include <limits.h>
  58 #include <signal.h>
  59 #include <stdio.h>
  60 #include <stdlib.h>
  61 #include <string.h>
  62 #include <unistd.h>
  63 #include "un-namespace.h"
  64
  65 #include <db.h>
  66 #include "btree.h"
  67
  68 #ifdef DEBUG
  69 #undef  MINPSIZE
  70 #define MINPSIZE        128
  71 #endif
  72
  73 static int byteorder(void);
  74 static int nroot(BTREE *);
  75 static int tmp(void);
  76
  77 /*
  78  * __BT_OPEN -- Open a btree.
  79  *
  80  * Creates and fills a DB struct, and calls the routine that actually
  81  * opens the btree.
  82  *
  83  * Parameters:
  84  *      fname:  filename (NULL for in-memory trees)
  85  *      flags:  open flag bits
  86  *      mode:   open permission bits
  87  *      b:      BTREEINFO pointer
  88  *
  89  * Returns:
  90  *      NULL on failure, pointer to DB on success.
  91  *
  92  */
  93 DB *
  94 __bt_open(fname, flags, mode, openinfo, dflags)
  95         const char *fname;
  96         int flags, mode, dflags;
  97         const BTREEINFO *openinfo;
  98 {
  99         struct stat sb;
 100         BTMETA m;
 101         BTREE *t;
 102         BTREEINFO b;
 103         DB *dbp;
 104         pgno_t ncache;
 105         ssize_t nr;
 106         int machine_lorder;
 107
 108         t = NULL;
 109
 110         /*
 111          * Intention is to make sure all of the user's selections are okay
 112          * here and then use them without checking.  Can't be complete, since
 113          * we don't know the right page size, lorder or flags until the backing
 114          * file is opened.  Also, the file's page size can cause the cachesize
 115          * to change.
 116          */
 117         machine_lorder = byteorder();
 118         if (openinfo) {
 119                 b = *openinfo;
 120
 121                 /* Flags: R_DUP. */
 122                 if (b.flags & ~(R_DUP))
 123                         goto einval;
 124
 125                 /*
 126                  * Page size must be indx_t aligned and >= MINPSIZE.  Default
 127                  * page size is set farther on, based on the underlying file
 128                  * transfer size.
 129                  */
 130                 if (b.psize &&
 131                     (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
 132                     b.psize & (sizeof(indx_t) - 1) ))
 133                         goto einval;
 134
 135                 /* Minimum number of keys per page; absolute minimum is 2. */
 136                 if (b.minkeypage) {
 137                         if (b.minkeypage < 2)
 138                                 goto einval;
 139                 } else
 140                         b.minkeypage = DEFMINKEYPAGE;
 141
 142                 /* If no comparison, use default comparison and prefix. */
 143                 if (b.compare == NULL) {
 144                         b.compare = __bt_defcmp;
 145                         if (b.prefix == NULL)
 146                                 b.prefix = __bt_defpfx;
 147                 }
 148
 149                 if (b.lorder == 0)
 150                         b.lorder = machine_lorder;
 151         } else {
 152                 b.compare = __bt_defcmp;
 153                 b.cachesize = 0;
 154                 b.flags = 0;
 155                 b.lorder = machine_lorder;
 156                 b.minkeypage = DEFMINKEYPAGE;
 157                 b.prefix = __bt_defpfx;
 158                 b.psize = 0;
 159         }
 160
 161         /* Check for the ubiquitous PDP-11. */
 162         if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
 163                 goto einval;
 164
 165         /* Allocate and initialize DB and BTREE structures. */
 166         if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
 167                 goto err;
 168         memset(t, 0, sizeof(BTREE));
 169         t->bt_fd = -1;                  /* Don't close unopened fd on error. */
 170         t->bt_lorder = b.lorder;
 171         t->bt_order = NOT;
 172         t->bt_cmp = b.compare;
 173         t->bt_pfx = b.prefix;
 174         t->bt_rfd = -1;
 175
 176         if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
 177                 goto err;
 178         memset(t->bt_dbp, 0, sizeof(DB));
 179         if (t->bt_lorder != machine_lorder)
 180                 F_SET(t, B_NEEDSWAP);
 181
 182         dbp->type = DB_BTREE;
 183         dbp->internal = t;
 184         dbp->close = __bt_close;
 185         dbp->del = __bt_delete;
 186         dbp->fd = __bt_fd;
 187         dbp->get = __bt_get;
 188         dbp->put = __bt_put;
 189         dbp->seq = __bt_seq;
 190         dbp->sync = __bt_sync;
 191
 192         /*
 193          * If no file name was supplied, this is an in-memory btree and we
 194          * open a backing temporary file.  Otherwise, it's a disk-based tree.
 195          */
 196         if (fname) {
 197                 switch (flags & O_ACCMODE) {
 198                 case O_RDONLY:
 199                         F_SET(t, B_RDONLY);
 200                         break;
 201                 case O_RDWR:
 202                         break;
 203                 case O_WRONLY:
 204                 default:
 205                         goto einval;
 206                 }
 207
 208                 if ((t->bt_fd = _open(fname, flags, mode)) < 0)
 209                         goto err;
 210
 211         } else {
 212                 if ((flags & O_ACCMODE) != O_RDWR)
 213                         goto einval;
 214                 if ((t->bt_fd = tmp()) == -1)
 215                         goto err;
 216                 F_SET(t, B_INMEM);
 217         }
 218
 219         if (_fcntl(t->bt_fd, F_SETFD, 1) == -1)
 220                 goto err;
 221
 222         if (_fstat(t->bt_fd, &sb))
 223                 goto err;
 224         if (sb.st_size) {
 225                 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
 226                         goto err;
 227                 if (nr != sizeof(BTMETA))
 228                         goto eftype;
 229
 230                 /*
 231                  * Read in the meta-data.  This can change the notion of what
 232                  * the lorder, page size and flags are, and, when the page size
 233                  * changes, the cachesize value can change too.  If the user
 234                  * specified the wrong byte order for an existing database, we
 235                  * don't bother to return an error, we just clear the NEEDSWAP
 236                  * bit.
 237                  */
 238                 if (m.magic == BTREEMAGIC)
 239                         F_CLR(t, B_NEEDSWAP);
 240                 else {
 241                         F_SET(t, B_NEEDSWAP);
 242                         M_32_SWAP(m.magic);
 243                         M_32_SWAP(m.version);
 244                         M_32_SWAP(m.psize);
 245                         M_32_SWAP(m.free);
 246                         M_32_SWAP(m.nrecs);
 247                         M_32_SWAP(m.flags);
 248                 }
 249                 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
 250                         goto eftype;
 251                 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
 252                     m.psize & (sizeof(indx_t) - 1) )
 253                         goto eftype;
 254                 if (m.flags & ~SAVEMETA)
 255                         goto eftype;
 256                 b.psize = m.psize;
 257                 F_SET(t, m.flags);
 258                 t->bt_free = m.free;
 259                 t->bt_nrecs = m.nrecs;
 260         } else {
 261                 /*
 262                  * Set the page size to the best value for I/O to this file.
 263                  * Don't overflow the page offset type.
 264                  */
 265                 if (b.psize == 0) {
 266                         b.psize = sb.st_blksize;
 267                         if (b.psize < MINPSIZE)
 268                                 b.psize = MINPSIZE;
 269                         if (b.psize > MAX_PAGE_OFFSET + 1)
 270                                 b.psize = MAX_PAGE_OFFSET + 1;
 271                 }
 272
 273                 /* Set flag if duplicates permitted. */
 274                 if (!(b.flags & R_DUP))
 275                         F_SET(t, B_NODUPS);
 276
 277                 t->bt_free = P_INVALID;
 278                 t->bt_nrecs = 0;
 279                 F_SET(t, B_METADIRTY);
 280         }
 281
 282         t->bt_psize = b.psize;
 283
 284         /* Set the cache size; must be a multiple of the page size. */
 285         if (b.cachesize && b.cachesize & (b.psize - 1) )
 286                 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1;
 287         if (b.cachesize < b.psize * MINCACHE)
 288                 b.cachesize = b.psize * MINCACHE;
 289
 290         /* Calculate number of pages to cache. */
 291         ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
 292
 293         /*
 294          * The btree data structure requires that at least two keys can fit on
 295          * a page, but other than that there's no fixed requirement.  The user
 296          * specified a minimum number per page, and we translated that into the
 297          * number of bytes a key/data pair can use before being placed on an
 298          * overflow page.  This calculation includes the page header, the size
 299          * of the index referencing the leaf item and the size of the leaf item
 300          * structure.  Also, don't let the user specify a minkeypage such that
 301          * a key/data pair won't fit even if both key and data are on overflow
 302          * pages.
 303          */
 304         t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
 305             (sizeof(indx_t) + NBLEAFDBT(0, 0));
 306         if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
 307                 t->bt_ovflsize =
 308                     NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
 309
 310         /* Initialize the buffer pool. */
 311         if ((t->bt_mp =
 312             mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
 313                 goto err;
 314         if (!F_ISSET(t, B_INMEM))
 315                 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
 316
 317         /* Create a root page if new tree. */
 318         if (nroot(t) == RET_ERROR)
 319                 goto err;
 320
 321         /* Global flags. */
 322         if (dflags & DB_LOCK)
 323                 F_SET(t, B_DB_LOCK);
 324         if (dflags & DB_SHMEM)
 325                 F_SET(t, B_DB_SHMEM);
 326         if (dflags & DB_TXN)
 327                 F_SET(t, B_DB_TXN);
 328
 329         return (dbp);
 330
 331 einval: errno = EINVAL;
 332         goto err;
 333
 334 eftype: errno = EFTYPE;
 335         goto err;
 336
 337 err:    if (t) {
 338                 if (t->bt_dbp)
 339                         free(t->bt_dbp);
 340                 if (t->bt_fd != -1)
 341                         (void)_close(t->bt_fd);
 342                 free(t);
 343         }
 344         return (NULL);
 345 }
 346
 347 /*
 348  * NROOT -- Create the root of a new tree.
 349  *
 350  * Parameters:
 351  *      t:      tree
 352  *
 353  * Returns:
 354  *      RET_ERROR, RET_SUCCESS
 355  */
 356 static int
 357 nroot(t)
 358         BTREE *t;
 359 {
 360         PAGE *meta, *root;
 361         pgno_t npg;
 362
 363         if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) {
 364                 mpool_put(t->bt_mp, meta, 0);
 365                 return (RET_SUCCESS);
 366         }
 367         if (errno != EINVAL)            /* It's OK to not exist. */
 368                 return (RET_ERROR);
 369         errno = 0;
 370
 371         if ((meta = mpool_new(t->bt_mp, &npg)) == NULL)
 372                 return (RET_ERROR);
 373
 374         if ((root = mpool_new(t->bt_mp, &npg)) == NULL)
 375                 return (RET_ERROR);
 376
 377         if (npg != P_ROOT)
 378                 return (RET_ERROR);
 379         root->pgno = npg;
 380         root->prevpg = root->nextpg = P_INVALID;
 381         root->lower = BTDATAOFF;
 382         root->upper = t->bt_psize;
 383         root->flags = P_BLEAF;
 384         memset(meta, 0, t->bt_psize);
 385         mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
 386         mpool_put(t->bt_mp, root, MPOOL_DIRTY);
 387         return (RET_SUCCESS);
 388 }
 389
 390 static int
 391 tmp()
 392 {
 393         sigset_t set, oset;
 394         int fd;
 395         char *envtmp = NULL;
 396         char path[MAXPATHLEN];
 397
 398         if (issetugid() == 0)
 399                 envtmp = getenv("TMPDIR");
 400         (void)snprintf(path,
 401             sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp");
 402
 403         (void)sigfillset(&set);
 404         (void)_sigprocmask(SIG_BLOCK, &set, &oset);
 405         if ((fd = mkstemp(path)) != -1)
 406                 (void)unlink(path);
 407         (void)_sigprocmask(SIG_SETMASK, &oset, NULL);
 408         return(fd);
 409 }
 410
 411 static int
 412 byteorder()
 413 {
 414         u_int32_t x;
 415         u_char *p;
 416
 417         x = 0x01020304;
 418         p = (u_char *)&x;
 419         switch (*p) {
 420         case 1:
 421                 return (BIG_ENDIAN);
 422         case 4:
 423                 return (LITTLE_ENDIAN);
 424         default:
 425                 return (0);
 426         }
 427 }
 428
 429 int
 430 __bt_fd(dbp)
 431         const DB *dbp;
 432 {
 433         BTREE *t;
 434
 435         t = dbp->internal;
 436
 437         /* Toss any page pinned across calls. */
 438         if (t->bt_pinned != NULL) {
 439                 mpool_put(t->bt_mp, t->bt_pinned, 0);
 440                 t->bt_pinned = NULL;
 441         }
 442
 443         /* In-memory database can't have a file descriptor. */
 444         if (F_ISSET(t, B_INMEM)) {
 445                 errno = ENOENT;
 446                 return (-1);
 447         }
 448         return (t->bt_fd);
 449 }